def run(self):
master_invoke = self.prm.master_invoke
launch_fn = os.path.join(master_invoke.network_dir,
self.remote_host) + '.smf_launch'
pickle_fn = master_invoke.host_result_filename(self.remote_host)
abortfn = master_invoke.abort_fn()
ensure_deleted(launch_fn)
ensure_deleted(pickle_fn)
if self.prm.master_invoke.verbose:
print('wrote command %s to launch file %s' % (self.remote_cmd, launch_fn))
write_sync_file(launch_fn, self.remote_cmd)
pickle_fn = master_invoke.host_result_filename(self.remote_host)
# print('waiting for pickle file %s'%pickle_fn)
self.status = master_invoke.NOTOK # premature exit means failure
while not os.path.exists(pickle_fn):
# print('%s not seen'%pickle_fn)
if os.path.exists(abortfn):
if master_invoke.verbose:
print('test abort seen by host ' + self.remote_host)
return
time.sleep(1.0)
self.status = master_invoke.OK # success!
def run_multi_thread_workload(prm):
master_invoke = prm.master_invoke
prm_slave = prm.is_slave
verbose = master_invoke.verbose
host = master_invoke.onhost
if not prm_slave:
sync_files.create_top_dirs(master_invoke, False)
if prm_slave:
time.sleep(1.1)
for d in master_invoke.top_dirs:
ensure_dir_exists(d)
for dlist in [master_invoke.src_dirs, master_invoke.dest_dirs]:
for d in dlist:
ensure_dir_exists(d)
if verbose:
print(host + ' saw ' + str(d))
# for each thread set up SmallfileWorkload instance,
# create a thread instance, and delete the thread-ready file
thread_list = create_worker_list(prm)
my_host_invoke = thread_list[0].invoke
# start threads, wait for them to reach starting gate
# to do this, look for thread-ready files
for t in thread_list:
ensure_deleted(t.invoke.gen_thread_ready_fname(t.invoke.tid))
for t in thread_list:
t.start()
if verbose:
print('started %d worker threads on host %s' %
(len(thread_list), host))
# wait for all threads to reach the starting gate
# this makes it more likely that they will start simultaneously
startup_timeout = prm.startup_timeout
if smallfile.is_windows_os:
print('adding time for Windows synchronization')
startup_timeout += 30
abort_fname = my_host_invoke.abort_fn()
thread_count = len(thread_list)
thread_to_wait_for = 0
for sec in range(0, startup_timeout * 2):
for k in range(thread_to_wait_for, thread_count):
t = thread_list[k]
fn = t.invoke.gen_thread_ready_fname(t.invoke.tid)
if not os.path.exists(fn):
if verbose:
print('thread %d thread-ready file %s not found...' %
(k, fn))
break
thread_to_wait_for = k + 1
if thread_to_wait_for == thread_count:
break
if os.path.exists(abort_fname):
break
time.sleep(0.5)
# if all threads didn't make it to the starting gate
if thread_to_wait_for < thread_count:
abort_test(abort_fname, thread_list)
raise Exception('only %d threads reached starting gate within %d sec' %
(thread_to_wait_for, startup_timeout))
# declare that this host is at the starting gate
if prm_slave:
host_ready_fn = my_host_invoke.gen_host_ready_fname()
if my_host_invoke.verbose:
print('host %s creating ready file %s' %
(my_host_invoke.onhost, host_ready_fn))
smallfile.touch(host_ready_fn)
sg = my_host_invoke.starting_gate
if not prm_slave: # special case of no --host-set parameter
try:
sync_files.write_sync_file(sg, 'hi there')
if verbose:
print('wrote starting gate file')
except IOError as e:
print('error writing starting gate for threads: %s' % str(e))
# wait for starting_gate file to be created by test driver
# every second we resume scan from last host file not found
if verbose:
print('awaiting ' + sg)
if prm_slave:
for sec in range(0, prm.host_startup_timeout + 10):
# hack to ensure that directory is up to date
# ndlist = os.listdir(my_host_invoke.network_dir)
# if verbose: print(str(ndlist))
if os.path.exists(sg):
break
time.sleep(0.5)
if not os.path.exists(sg):
abort_test(my_host_invoke.abort_fn(), thread_list)
raise Exception('starting signal not seen within %d seconds' %
prm.host_startup_timeout)
if verbose:
print('starting test on host ' + host + ' in 2 seconds')
time.sleep(2 + random.random()) # let other hosts see starting gate file
# FIXME: don't timeout the test,
# instead check thread progress and abort if you see any of them stalled
# but if servers are heavily loaded you can't rely on filesystem
# wait for all threads on this host to finish
for t in thread_list:
if verbose:
print('waiting for thread %s' % t.invoke.tid)
t.invoke = t.receiver.recv() # to get results from sub-process
t.join()
# if not a slave of some other host, print results (for this host)
exit_status = OK
if not prm_slave:
try:
invoke_list = [t.invoke for t in thread_list]
output_results.output_results(invoke_list, prm)
except SMFResultException as e:
print('ERROR: ' + str(e))
exit_status = NOTOK
else:
# if we are participating in a multi-host test
# then write out this host's result in pickle format
# so test driver can pick up result
result_filename = \
master_invoke.host_result_filename(prm.as_host)
if verbose:
print('writing invokes to: ' + result_filename)
invok_list = [t.invoke for t in thread_list]
if verbose:
print('saving result to filename %s' % result_filename)
for ivk in invok_list:
ivk.buf = None
ivk.biggest_buf = None
sync_files.write_pickle(result_filename, invok_list)
time.sleep(1.2) # for benefit of NFS with actimeo=1
sys.exit(exit_status)
def run_multi_thread_workload(prm):
master_invoke = prm.master_invoke
prm_slave = prm.is_slave
verbose = master_invoke.verbose
host = master_invoke.onhost
if not prm_slave:
sync_files.create_top_dirs(master_invoke, False)
if prm_slave:
time.sleep(1.1)
for d in master_invoke.top_dirs:
ensure_dir_exists(d)
for dlist in [master_invoke.src_dirs, master_invoke.dest_dirs]:
for d in dlist:
ensure_dir_exists(d)
if verbose:
print(host + ' saw ' + str(d))
# for each thread set up SmallfileWorkload instance,
# create a thread instance, and delete the thread-ready file
thread_list = create_worker_list(prm)
my_host_invoke = thread_list[0].invoke
# start threads, wait for them to reach starting gate
# to do this, look for thread-ready files
for t in thread_list:
ensure_deleted(t.invoke.gen_thread_ready_fname(t.invoke.tid))
for t in thread_list:
t.start()
if verbose:
print('started %d worker threads on host %s' %
(len(thread_list), host))
# wait for all threads to reach the starting gate
# this makes it more likely that they will start simultaneously
startup_timeout = prm.startup_timeout
if smallfile.is_windows_os:
print('adding time for Windows synchronization')
startup_timeout += 30
abort_fname = my_host_invoke.abort_fn()
thread_count = len(thread_list)
thread_to_wait_for = 0
for sec in range(0, startup_timeout * 2):
for k in range(thread_to_wait_for, thread_count):
t = thread_list[k]
fn = t.invoke.gen_thread_ready_fname(t.invoke.tid)
if not os.path.exists(fn):
if verbose:
print('thread %d thread-ready file %s not found...' % (k, fn))
break
thread_to_wait_for = k + 1
if thread_to_wait_for == thread_count:
break
if os.path.exists(abort_fname):
break
time.sleep(0.5)
# if all threads didn't make it to the starting gate
if thread_to_wait_for < thread_count:
abort_test(abort_fname, thread_list)
raise Exception('only %d threads reached starting gate within %d sec'
% (thread_to_wait_for, startup_timeout))
# declare that this host is at the starting gate
if prm_slave:
host_ready_fn = my_host_invoke.gen_host_ready_fname()
if my_host_invoke.verbose:
print('host %s creating ready file %s' %
(my_host_invoke.onhost, host_ready_fn))
smallfile.touch(host_ready_fn)
sg = my_host_invoke.starting_gate
if not prm_slave: # special case of no --host-set parameter
try:
sync_files.write_sync_file(sg, 'hi there')
if verbose:
print('wrote starting gate file')
except IOError as e:
print('error writing starting gate for threads: %s' % str(e))
# wait for starting_gate file to be created by test driver
# every second we resume scan from last host file not found
if verbose:
print('awaiting ' + sg)
if prm_slave:
for sec in range(0, prm.host_startup_timeout + 10):
# hack to ensure that directory is up to date
# ndlist = os.listdir(my_host_invoke.network_dir)
# if verbose: print(str(ndlist))
if os.path.exists(sg):
break
time.sleep(0.5)
if not os.path.exists(sg):
abort_test(my_host_invoke.abort_fn(), thread_list)
raise Exception('starting signal not seen within %d seconds'
% prm.host_startup_timeout)
if verbose:
print('starting test on host ' + host + ' in 2 seconds')
time.sleep(2 + random.random()) # let other hosts see starting gate file
# FIXME: don't timeout the test,
# instead check thread progress and abort if you see any of them stalled
# but if servers are heavily loaded you can't rely on filesystem
# wait for all threads on this host to finish
for t in thread_list:
if verbose:
print('waiting for thread %s' % t.invoke.tid)
t.invoke = t.receiver.recv() # to get results from sub-process
t.join()
# if not a slave of some other host, print results (for this host)
exit_status = OK
if not prm_slave:
try:
invoke_list = [t.invoke for t in thread_list]
output_results.output_results(invoke_list, prm)
except SMFResultException as e:
print('ERROR: ' + str(e))
exit_status = NOTOK
else:
# if we are participating in a multi-host test
# then write out this host's result in pickle format
# so test driver can pick up result
result_filename = \
master_invoke.host_result_filename(prm.as_host)
if verbose:
print('writing invokes to: ' + result_filename)
invok_list = [t.invoke for t in thread_list]
if verbose:
print('saving result to filename %s' % result_filename)
for ivk in invok_list:
ivk.buf = None
ivk.biggest_buf = None
sync_files.write_pickle(result_filename, invok_list)
time.sleep(1.2) # for benefit of NFS with actimeo=1
sys.exit(exit_status)
def run_multi_host_workload(prm):
prm_host_set = prm.host_set
prm_permute_host_dirs = prm.permute_host_dirs
master_invoke = prm.master_invoke
starting_gate = master_invoke.starting_gate
verbose = master_invoke.verbose
# construct list of ssh threads to invoke in parallel
sync_files.create_top_dirs(master_invoke, True)
pickle_fn = os.path.join(prm.master_invoke.network_dir, 'param.pickle')
# if verbose: print('writing ' + pickle_fn))
sync_files.write_pickle(pickle_fn, prm)
if os.getenv('PYPY'):
python_prog = os.getenv('PYPY')
elif sys.version.startswith('2'):
python_prog = 'python'
elif sys.version.startswith('3'):
python_prog = 'python3'
else:
raise Exception('unrecognized python version %s' % sys.version)
# print('python_prog = %s'%python_prog)
remote_thread_list = []
host_ct = len(prm_host_set)
for j in range(0, len(prm_host_set)):
remote_host = prm_host_set[j]
smf_remote_pgm = os.path.join(prm.remote_pgm_dir,
'smallfile_remote.py')
this_remote_cmd = '%s %s --network-sync-dir %s ' \
% (python_prog, smf_remote_pgm, prm.master_invoke.network_dir)
# this_remote_cmd = remote_cmd
if prm_permute_host_dirs:
this_remote_cmd += \
' --as-host %s' % prm_host_set[(j + 1) % host_ct]
else:
this_remote_cmd += ' --as-host %s' % remote_host
if verbose:
print(this_remote_cmd)
if smallfile.is_windows_os or prm.launch_by_daemon:
remote_thread_list.append(
launcher_thread.launcher_thread(prm, remote_host,
this_remote_cmd))
else:
remote_thread_list.append(
ssh_thread.ssh_thread(remote_host, this_remote_cmd))
# start them
for t in remote_thread_list:
if not prm.launch_by_daemon:
# pace starts so that we don't get ssh errors
time.sleep(0.1)
t.start()
# wait for hosts to arrive at starting gate
# if only one host, then no wait will occur
# as starting gate file is already present
# every second we resume scan from last host file not found
# FIXME: for very large host sets,
# timeout only if no host responds within X seconds
exception_seen = None
hosts_ready = False # set scope outside while loop
abortfn = master_invoke.abort_fn()
last_host_seen = -1
sec = 0.0
sec_delta = 0.5
host_timeout = prm.host_startup_timeout
if smallfile.is_windows_os:
host_timeout += 20
try:
while sec < host_timeout:
# HACK to force directory entry coherency for Gluster
ndirlist = os.listdir(master_invoke.network_dir)
if master_invoke.verbose:
print('shared dir list: ' + str(ndirlist))
hosts_ready = True
if os.path.exists(abortfn):
raise Exception('worker host signaled abort')
for j in range(last_host_seen + 1, len(prm_host_set)):
h = prm_host_set[j]
fn = master_invoke.gen_host_ready_fname(h.strip())
if verbose:
print('checking for host filename ' + fn)
if not os.path.exists(fn):
hosts_ready = False
break
last_host_seen = j # saw this host's ready file
# we exit while loop only if no hosts in host_timeout seconds
sec = 0.0
if hosts_ready:
break
# if one of ssh threads has died, no reason to continue
kill_remaining_threads = False
for t in remote_thread_list:
if not t.isAlive():
print('thread %s on host %s has died' % (t, h))
kill_remaining_threads = True
break
if kill_remaining_threads:
break
# be patient for large tests
# give user some feedback about
# how many hosts have arrived at the starting gate
time.sleep(sec_delta)
sec += sec_delta
sec_delta += 1
if verbose:
print('last_host_seen=%d sec=%d' % (last_host_seen, sec))
except KeyboardInterrupt as e:
print('saw SIGINT signal, aborting test')
exception_seen = e
except Exception as e:
exception_seen = e
hosts_ready = False
if not hosts_ready:
smallfile.abort_test(abortfn, [])
print('ERROR: host %s did not reach starting gate' % h)
if not exception_seen:
raise Exception('hosts did not reach starting gate ' +
'within %d seconds' % host_timeout)
else:
print('saw exception %s, aborting test' % str(e))
else:
# ask all hosts to start the test
# this is like firing the gun at the track meet
try:
sync_files.write_sync_file(starting_gate, 'hi')
print('starting all threads by creating starting gate file %s' %
starting_gate)
except IOError as e:
print('error writing starting gate: %s' % os.strerror(e.errno))
# wait for them to finish
for t in remote_thread_list:
t.join()
if t.status != OK:
print('ERROR: ssh thread for host %s completed with status %d' %
(t.remote_host, t.status))
# attempt to aggregate results by reading pickle files
# containing SmallfileWorkload instances
# with counters and times that we need
try:
all_ok = NOTOK
invoke_list = []
one_shot_delay = True
for h in prm_host_set: # for each host in test
# read results for each thread run in that host
# from python pickle of the list of SmallfileWorkload objects
pickle_fn = master_invoke.host_result_filename(h)
if verbose:
print('reading pickle file: %s' % pickle_fn)
host_invoke_list = []
try:
if one_shot_delay and not os.path.exists(pickle_fn):
# all threads have joined already, they are done
# we allow > 1 sec
# for this (NFS) client to see other clients' files
time.sleep(1.2)
one_shot_delay = False
with open(pickle_fn, 'rb') as pickle_file:
host_invoke_list = pickle.load(pickle_file)
if verbose:
print(' read %d invoke objects' % len(host_invoke_list))
invoke_list.extend(host_invoke_list)
ensure_deleted(pickle_fn)
except IOError as e:
if e.errno != errno.ENOENT:
raise e
print(' pickle file %s not found' % pickle_fn)
output_results.output_results(invoke_list, prm)
all_ok = OK
except IOError as e:
print('host %s filename %s: %s' % (h, pickle_fn, str(e)))
except KeyboardInterrupt as e:
print('control-C signal seen (SIGINT)')
except SMFResultException as e:
print(str(e))
sys.exit(all_ok)
def run_multi_host_workload(prm, log):
# construct list of ssh threads to invoke in parallel
if os.getenv('PYPY'):
python_prog = os.getenv('PYPY')
elif sys.version.startswith('2'):
python_prog = 'python'
elif sys.version.startswith('3'):
python_prog = 'python3'
else:
raise Exception('unrecognized python version %s' % sys.version)
log.debug('python_prog = %s' % python_prog)
remote_thread_list = []
host_ct = len(prm.host_set)
for j in range(0, len(prm.host_set)):
remote_host = prm.host_set[j]
fsd_remote_pgm = os.path.join(prm.fsd_remote_dir, 'fs-drift-remote.py')
this_remote_cmd = '%s %s --network-sync-dir %s ' \
% (prm.python_prog, fsd_remote_pgm, prm.network_shared_path)
this_remote_cmd += ' --as-host %s' % remote_host
log.debug(this_remote_cmd)
if prm.launch_as_daemon:
remote_thread_list.append(
launcher_thread.launcher_thread(prm, log, remote_host,
this_remote_cmd))
else:
remote_thread_list.append(
ssh_thread.ssh_thread(log, remote_host, this_remote_cmd))
# start them, pacing starts so that we don't get ssh errors
for t in remote_thread_list:
if prm.launch_as_daemon:
time.sleep(0.1)
t.start()
# wait for hosts to arrive at starting gate
# if only one host, then no wait will occur
# as starting gate file is already present
# every second we resume scan from last host file not found
exception_seen = None
abortfn = prm.abort_path
sec_delta = 0.5
# timeout if no host replies in next host_timeout seconds
per_host_timeout = 10.0
all_host_timeout = 5.0 + len(prm.host_set) / 3
if all_host_timeout < per_host_timeout:
per_host_timeout = all_host_timeout / 2
hosts_ready = False # set scope outside while loop
last_host_seen = -1
sec = 0.0
start_loop_start = time.time()
try:
while sec < per_host_timeout:
# HACK to force directory entry coherency for Gluster
#ndirlist = os.listdir(prm.network_shared_path)
#log.debug('shared dir list: ' + str(ndirlist))
hosts_ready = True
if os.path.exists(abortfn):
raise FsDriftException('worker host signaled abort')
for j in range(last_host_seen + 1, len(prm.host_set)):
h = prm.host_set[j]
fn = multi_thread_workload.gen_host_ready_fname(prm, h.strip())
log.debug('checking for host filename ' + fn)
if not os.path.exists(fn):
log.info('did not see host filename %s after %f sec' %
(fn, sec))
hosts_ready = False
break
log.debug('saw host filename ' + fn)
last_host_seen = j # saw this host's ready file
# we exit while loop only if no hosts in per_host_timeout seconds
sec = 0.0
if hosts_ready:
break
# if one of ssh threads has died, no reason to continue
kill_remaining_threads = False
for t in remote_thread_list:
if not t.is_alive():
log.error('thread %s has died' % t)
kill_remaining_threads = True
break
if kill_remaining_threads:
break
# be patient for large tests
# give user some feedback about
# how many hosts have arrived at the starting gate
time.sleep(sec_delta)
sec += sec_delta
time_since_loop_start = time.time() - start_loop_start
log.debug('last_host_seen=%d sec=%d' % (last_host_seen, sec))
if time_since_loop_start > all_host_timeout:
kill_remaining_threads = True
break
except KeyboardInterrupt as e:
log.error('saw SIGINT signal, aborting test')
exception_seen = e
except Exception as e:
exception_seen = e
log.exception(e)
hosts_ready = False
if not hosts_ready:
multi_thread_workload.abort_test(prm.abort_path, remote_thread_list)
if not exception_seen:
log.info(
'no additional hosts reached starting gate within %5.1f seconds'
% per_host_timeout)
return NOTOK
else:
raise exception_seen
else:
# ask all hosts to start the test
# this is like firing the gun at the track meet
try:
sync_files.write_sync_file(prm.starting_gun_path, 'hi')
log.debug('starting all threads by creating starting gun file %s' %
prm.starting_gun_path)
except IOError as e:
log.error('error writing starting gun file: %s' %
os.strerror(e.errno))
multi_thread_workload.abort_test(prm.abort_path,
remote_thread_list)
raise e
# wait for them to finish
for t in remote_thread_list:
t.join()
if t.status != OK:
log.error('ssh thread for host %s completed with status %d' %
(t.remote_host, t.status))
# attempt to aggregate results by reading pickle files
# containing SmallfileWorkload instances
# with counters and times that we need
try:
invoke_list = []
one_shot_delay = True
for h in prm.host_set: # for each host in test
# read results for each thread run in that host
# from python pickle of the list of SmallfileWorkload objects
pickle_fn = multi_thread_workload.host_result_filename(prm, h)
log.debug('reading pickle file: %s' % pickle_fn)
host_invoke_list = []
try:
if one_shot_delay and not os.path.exists(pickle_fn):
# all threads have joined already, they are done
# we allow > 1 sec
# for this (NFS) client to see other clients' files
time.sleep(1.2)
one_shot_delay = False
host_invoke_list = read_pickle(pickle_fn)
log.debug(' read %d invoke objects' % len(host_invoke_list))
invoke_list.extend(host_invoke_list)
ensure_deleted(pickle_fn)
except IOError as e:
if e.errno != errno.ENOENT:
raise e
log.error(' pickle file %s not found' % pickle_fn)
output_results.output_results(prm, invoke_list)
except IOError as e:
log.exception(e)
log.error('host %s filename %s: %s' % (h, pickle_fn, str(e)))
return NOTOK
except KeyboardInterrupt as e:
log.error('control-C signal seen (SIGINT)')
return NOTOK
except FsDriftException as e:
log.exception(e)
return NOTOK
return (OK)
def run_multi_host_workload(prm):
prm_host_set = prm.host_set
prm_permute_host_dirs = prm.permute_host_dirs
master_invoke = prm.master_invoke
starting_gate = master_invoke.starting_gate
verbose = master_invoke.verbose
# construct list of ssh threads to invoke in parallel
sync_files.create_top_dirs(master_invoke, True)
pickle_fn = os.path.join(prm.master_invoke.network_dir, 'param.pickle')
# if verbose: print('writing ' + pickle_fn))
sync_files.write_pickle(pickle_fn, prm)
if os.getenv('PYPY'):
python_prog = os.getenv('PYPY')
elif sys.version.startswith('2'):
python_prog = 'python'
elif sys.version.startswith('3'):
python_prog = 'python3'
else:
raise Exception('unrecognized python version %s' % sys.version)
# print('python_prog = %s'%python_prog)
remote_thread_list = []
host_ct = len(prm_host_set)
for j in range(0, len(prm_host_set)):
remote_host = prm_host_set[j]
smf_remote_pgm = os.path.join(prm.remote_pgm_dir,
'smallfile_remote.py')
this_remote_cmd = '%s %s --network-sync-dir %s ' \
% (python_prog, smf_remote_pgm, prm.master_invoke.network_dir)
# this_remote_cmd = remote_cmd
if prm_permute_host_dirs:
this_remote_cmd += \
' --as-host %s' % prm_host_set[(j + 1) % host_ct]
else:
this_remote_cmd += ' --as-host %s' % remote_host
if verbose:
print(this_remote_cmd)
if smallfile.is_windows_os or prm.launch_by_daemon:
remote_thread_list.append(
launcher_thread.launcher_thread(prm,
remote_host,
this_remote_cmd))
else:
remote_thread_list.append(ssh_thread.ssh_thread(remote_host,
this_remote_cmd))
# start them
for t in remote_thread_list:
if not prm.launch_by_daemon:
# pace starts so that we don't get ssh errors
time.sleep(0.1)
t.start()
# wait for hosts to arrive at starting gate
# if only one host, then no wait will occur
# as starting gate file is already present
# every second we resume scan from last host file not found
# FIXME: for very large host sets,
# timeout only if no host responds within X seconds
exception_seen = None
hosts_ready = False # set scope outside while loop
abortfn = master_invoke.abort_fn()
last_host_seen = -1
sec = 0.0
sec_delta = 0.5
host_timeout = prm.host_startup_timeout
if smallfile.is_windows_os:
host_timeout += 20
try:
while sec < host_timeout:
# HACK to force directory entry coherency for Gluster
ndirlist = os.listdir(master_invoke.network_dir)
if master_invoke.verbose:
print('shared dir list: ' + str(ndirlist))
hosts_ready = True
if os.path.exists(abortfn):
raise Exception('worker host signaled abort')
for j in range(last_host_seen + 1, len(prm_host_set)):
h = prm_host_set[j]
fn = master_invoke.gen_host_ready_fname(h.strip())
if verbose:
print('checking for host filename ' + fn)
if not os.path.exists(fn):
hosts_ready = False
break
last_host_seen = j # saw this host's ready file
# we exit while loop only if no hosts in host_timeout seconds
sec = 0.0
if hosts_ready:
break
# if one of ssh threads has died, no reason to continue
kill_remaining_threads = False
for t in remote_thread_list:
if not t.isAlive():
print('thread %s on host %s has died' % (t, h))
kill_remaining_threads = True
break
if kill_remaining_threads:
break
# be patient for large tests
# give user some feedback about
# how many hosts have arrived at the starting gate
time.sleep(sec_delta)
sec += sec_delta
sec_delta += 1
if verbose:
print('last_host_seen=%d sec=%d' % (last_host_seen, sec))
except KeyboardInterrupt as e:
print('saw SIGINT signal, aborting test')
exception_seen = e
except Exception as e:
exception_seen = e
hosts_ready = False
if not hosts_ready:
smallfile.abort_test(abortfn, [])
print('ERROR: host %s did not reach starting gate' % h)
if not exception_seen:
raise Exception('hosts did not reach starting gate ' +
'within %d seconds' % host_timeout)
else:
print('saw exception %s, aborting test' % str(e))
else:
# ask all hosts to start the test
# this is like firing the gun at the track meet
try:
sync_files.write_sync_file(starting_gate, 'hi')
print('starting all threads by creating starting gate file %s' %
starting_gate)
except IOError as e:
print('error writing starting gate: %s' % os.strerror(e.errno))
# wait for them to finish
for t in remote_thread_list:
t.join()
if t.status != OK:
print('ERROR: ssh thread for host %s completed with status %d' %
(t.remote_host, t.status))
# attempt to aggregate results by reading pickle files
# containing SmallfileWorkload instances
# with counters and times that we need
try:
all_ok = NOTOK
invoke_list = []
one_shot_delay = True
for h in prm_host_set: # for each host in test
# read results for each thread run in that host
# from python pickle of the list of SmallfileWorkload objects
pickle_fn = master_invoke.host_result_filename(h)
if verbose:
print('reading pickle file: %s' % pickle_fn)
host_invoke_list = []
try:
if one_shot_delay and not os.path.exists(pickle_fn):
# all threads have joined already, they are done
# we allow > 1 sec
# for this (NFS) client to see other clients' files
time.sleep(1.2)
one_shot_delay = False
with open(pickle_fn, 'rb') as pickle_file:
host_invoke_list = pickle.load(pickle_file)
if verbose:
print(' read %d invoke objects' % len(host_invoke_list))
invoke_list.extend(host_invoke_list)
ensure_deleted(pickle_fn)
except IOError as e:
if e.errno != errno.ENOENT:
raise e
print(' pickle file %s not found' % pickle_fn)
output_results.output_results(invoke_list, prm)
all_ok = OK
except IOError as e:
print('host %s filename %s: %s' % (h, pickle_fn, str(e)))
except KeyboardInterrupt as e:
print('control-C signal seen (SIGINT)')
except SMFResultException as e:
print(str(e))
sys.exit(all_ok)
def terminate(self):
sync_files.write_sync_file(self.params.abort_path, 'shut it down')
self.status = NOTOK
def run_multi_host_workload(prm):
prm_host_set = prm.host_set
prm_slave = prm.is_slave
prm_permute_host_dirs = prm.permute_host_dirs
master_invoke = prm.master_invoke
starting_gate = master_invoke.starting_gate
verbose = master_invoke.verbose
host = master_invoke.onhost
# construct list of ssh threads to invoke in parallel
sync_files.create_top_dirs(master_invoke, True)
pickle_fn = os.path.join(prm.master_invoke.network_dir,'param.pickle')
#if verbose: print('writing ' + pickle_fn)
sync_files.write_pickle(pickle_fn, prm)
if os.getenv('PYPY'):
python_prog = os.getenv('PYPY')
elif sys.version.startswith('2'):
python_prog = 'python'
elif sys.version.startswith('3'):
python_prog = 'python3'
else:
raise Exception('unrecognized python version %s'%sys.version)
#print('python_prog = %s'%python_prog)
remote_thread_list = []
host_ct = len(prm_host_set)
for j in range(0, len(prm_host_set)):
remote_host = prm_host_set[j]
smf_remote_pgm = os.path.join(prm.remote_pgm_dir, 'smallfile_remote.py')
this_remote_cmd = '%s %s --network-sync-dir %s '%\
(python_prog, smf_remote_pgm, prm.master_invoke.network_dir)
#this_remote_cmd = remote_cmd
if prm_permute_host_dirs:
this_remote_cmd += ' --as-host %s'%prm_host_set[(j+1)%host_ct]
else:
this_remote_cmd += ' --as-host %s'%remote_host
if verbose: print(this_remote_cmd)
if smallfile.is_windows_os:
remote_thread_list.append(launcher_thread.launcher_thread(prm, remote_host, this_remote_cmd ))
else:
remote_thread_list.append(ssh_thread.ssh_thread(remote_host, this_remote_cmd))
# start them, pacing starts so that we don't get ssh errors
for t in remote_thread_list:
t.start()
# wait for hosts to arrive at starting gate
# if only one host, then no wait will occur as starting gate file is already present
# every second we resume scan from last host file not found
# FIXME: for very large host sets, timeout only if no host responds within X seconds
exception_seen = None
hosts_ready = False # set scope outside while loop
abortfn = master_invoke.abort_fn()
last_host_seen=-1
sec = 0
sec_delta = 0.5
host_timeout = prm.host_startup_timeout
if smallfile.is_windows_os: host_timeout += 20
try:
# FIXME: make timeout criteria be that new new hosts seen in X seconds
while sec < host_timeout:
ndirlist = os.listdir(master_invoke.network_dir)
if master_invoke.verbose: print('shared dir list: ' + str(ndirlist))
hosts_ready = True
if os.path.exists(abortfn): raise Exception('worker host signaled abort')
for j in range(last_host_seen+1, len(prm_host_set)):
h=prm_host_set[j]
fn = master_invoke.gen_host_ready_fname(h.strip())
if verbose: print('checking for host filename '+fn)
if not os.path.exists(fn):
hosts_ready = False
break
last_host_seen=j
if hosts_ready: break
# be patient for large tests
# give user some feedback about how many hosts have arrived at the starting gate
time.sleep(sec_delta)
sec += sec_delta
sec_delta += 1
if verbose: print('last_host_seen=%d sec=%d'%(last_host_seen,sec))
except KeyboardInterrupt as e:
print('saw SIGINT signal, aborting test')
exception_seen = e
except Exception as e:
exception_seen = e
hosts_ready = False
if not hosts_ready:
smallfile.abort_test(abortfn, [])
if not exception_seen:
raise Exception('hosts did not reach starting gate within %d seconds'%host_timeout)
else:
print('saw exception %s, aborting test'%str(e))
else:
# ask all hosts to start the test
# this is like firing the gun at the track meet
try:
sync_files.write_sync_file(starting_gate, 'hi')
if verbose: print('starting gate file %s created'%starting_gate)
except IOError as e:
print('error writing starting gate: %s'%os.strerror(e.errno))
# wait for them to finish
all_ok = True
for t in remote_thread_list:
t.join()
if t.status != OK:
all_ok = False
print('ERROR: ssh thread for host %s completed with status %d'%(t.remote_host, t.status))
# attempt to aggregate results by reading pickle files
# containing smf_invocation instances with counters and times that we need
try:
invoke_list = []
for h in prm_host_set: # for each host in test
# read results for each thread run in that host
# from python pickle of the list of smf_invocation objects
pickle_fn = master_invoke.host_result_filename(h)
if verbose: print('reading pickle file: %s'%pickle_fn)
host_invoke_list = []
try:
if not os.path.exists(pickle_fn): time.sleep(1.2)
with open(pickle_fn, 'rb') as pickle_file:
host_invoke_list = pickle.load(pickle_file)
if verbose: print(' read %d invoke objects'%len(host_invoke_list))
invoke_list.extend(host_invoke_list)
ensure_deleted(pickle_fn)
except IOError as e:
if e.errno != errno.ENOENT: raise e
print(' pickle file %s not found'%pickle_fn)
output_results.output_results(invoke_list, prm_host_set, prm.thread_count,pct_files_min)
except IOError as e:
print('host %s filename %s: %s'%(h, pickle_fn, str(e)))
all_ok = False
except KeyboardInterrupt as e:
print('control-C signal seen (SIGINT)')
all_ok = False
if not all_ok:
sys.exit(NOTOK)
sys.exit(OK)
def run_multi_thread_workload(prm):
master_invoke = prm.master_invoke
prm_slave = prm.is_slave
verbose = master_invoke.verbose
host = master_invoke.onhost
if not prm_slave:
sync_files.create_top_dirs(master_invoke, False)
if prm_slave:
time.sleep(1.1)
os.listdir(master_invoke.network_dir)
for dlist in [master_invoke.src_dirs, master_invoke.dest_dirs]:
for d in dlist:
os.listdir(d) # hack to ensure that
if verbose:
print(host + " saw " + d)
# for each thread set up smf_invocation instance,
# create a thread instance, and delete the thread-ready file
thread_list = create_worker_list(prm)
starting_gate = thread_list[0].invoke.starting_gate
my_host_invoke = thread_list[0].invoke
# start threads, wait for them to reach starting gate
# to do this, look for thread-ready files
for t in thread_list:
ensure_deleted(t.invoke.gen_thread_ready_fname(t.invoke.tid))
for t in thread_list:
t.start()
if verbose:
print("started %d worker threads on host %s" % (len(thread_list), host))
# wait for all threads to reach the starting gate
# this makes it more likely that they will start simultaneously
abort_fname = my_host_invoke.abort_fn()
threads_ready = False # really just to set scope of variable
k = 0
for sec in range(0, prm.startup_timeout * 2):
threads_ready = True
for t in thread_list:
fn = t.invoke.gen_thread_ready_fname(t.invoke.tid)
if not os.path.exists(fn):
threads_ready = False
break
if threads_ready:
break
if os.path.exists(abort_fname):
break
if verbose:
print("threads not ready...")
time.sleep(0.5)
# if all threads didn't make it to the starting gate
if not threads_ready:
abort_test(abort_fname, thread_list)
raise Exception("threads did not reach starting gate within %d sec" % prm.startup_timeout)
# declare that this host is at the starting gate
if prm_slave:
host_ready_fn = my_host_invoke.gen_host_ready_fname()
if my_host_invoke.verbose:
print("host %s creating ready file %s" % (my_host_invoke.onhost, host_ready_fn))
smallfile.touch(host_ready_fn)
sg = my_host_invoke.starting_gate
if not prm_slave: # special case of no --host-set parameter
try:
sync_files.write_sync_file(sg, "hi there")
if verbose:
print("wrote starting gate file")
except IOError as e:
print("error writing starting gate for threads: %s" % str(e))
# wait for starting_gate file to be created by test driver
# every second we resume scan from last host file not found
if verbose:
print("awaiting " + sg)
if prm_slave:
for sec in range(0, prm.host_startup_timeout * 2):
if os.path.exists(sg):
break
time.sleep(0.5)
if not os.path.exists(sg):
abort_test(my_host_invoke.abort_fn(), thread_list)
raise Exception("starting signal not seen within %d seconds" % prm.host_startup_timeout)
if verbose:
print("starting test on host " + host + " in 2 seconds")
time.sleep(2 + random.random())
# FIXME: don't timeout the test,
# instead check thread progress and abort if you see any of them stalled
# for long enough
# problem is: if servers are heavily loaded you can't use filesystem to communicate this
# wait for all threads on this host to finish
for t in thread_list:
if verbose:
print("waiting for thread %s" % t.invoke.tid)
t.invoke = t.receiver.recv() # must do this to get results from sub-process
t.join()
# if not a slave of some other host, print results (for this host)
exit_status = OK
if not prm_slave:
try:
# FIXME: code to aggregate results from list of invoke objects can be shared by multi-host and single-host cases
invoke_list = [t.invoke for t in thread_list]
output_results.output_results(invoke_list, ["localhost"], prm.thread_count, smallfile.pct_files_min)
except SMFResultException as e:
print("ERROR: " + str(e))
exit_status = NOTOK
else:
# if we are participating in a multi-host test
# then write out this host's result in pickle format so test driver can pick up result
result_filename = master_invoke.host_result_filename(prm.as_host)
if verbose:
print("writing invokes to: " + result_filename)
invok_list = [t.invoke for t in thread_list]
if verbose:
print("saving result to filename %s" % result_filename)
for ivk in invok_list:
ivk.buf = None
ivk.biggest_buf = None
sync_files.write_pickle(result_filename, invok_list)
time.sleep(1.2) # for benefit of NFS with actimeo=1
sys.exit(exit_status)
def run_multi_thread_workload(prm):
host = prm.as_host
if host == None:
host = 'localhost'
prm_slave = (prm.host_set != [])
# FIXME: get coherent logging level interface
verbose = os.getenv('LOGLEVEL_DEBUG' != None)
host_startup_timeout = 5 + len(prm.host_set) / 3
# for each thread set up SmallfileWorkload instance,
# create a thread instance, and delete the thread-ready file
thread_list = create_worker_list(prm)
my_host_invoke = thread_list[0].invoke
my_log = fsd_log.start_log('%s.master' % host)
my_log.debug(prm)
# start threads, wait for them to reach starting gate
# to do this, look for thread-ready files
for t in thread_list:
ensure_deleted(t.invoke.gen_thread_ready_fname(t.invoke.tid))
for t in thread_list:
t.start()
my_log.debug('started %d worker threads on host %s' %
(len(thread_list), host))
# wait for all threads to reach the starting gate
# this makes it more likely that they will start simultaneously
abort_fname = prm.abort_path
thread_count = len(thread_list)
thread_to_wait_for = 0
startup_timeout = 3
sec = 0.0
while sec < startup_timeout:
for k in range(thread_to_wait_for, thread_count):
t = thread_list[k]
fn = t.invoke.gen_thread_ready_fname(t.invoke.tid)
if not os.path.exists(fn):
my_log.debug('thread %d thread-ready file %s not found yet with %f sec left' %
(k, fn, (startup_timeout - sec)))
break
thread_to_wait_for = k + 1
# we only timeout if no more threads have reached starting gate
# in startup_timeout sec
sec = 0.0
if thread_to_wait_for == thread_count:
break
if os.path.exists(abort_fname):
break
sec += 0.5
time.sleep(0.5)
# if all threads didn't make it to the starting gate
if thread_to_wait_for < thread_count:
abort_test(abort_fname, thread_list)
raise FsDriftException('only %d threads reached starting gate'
% thread_to_wait_for)
# declare that this host is at the starting gate
if prm_slave:
host_ready_fn = gen_host_ready_fname(prm, prm.as_host)
my_log.debug('host %s creating ready file %s' %
(my_host_invoke.onhost, host_ready_fn))
common.touch(host_ready_fn)
sg = prm.starting_gun_path
if not prm_slave:
my_log.debug('wrote starting gate file ')
sync_files.write_sync_file(sg, 'hi there')
# wait for starting_gate file to be created by test driver
# every second we resume scan from last host file not found
if prm_slave:
my_log.debug('awaiting ' + sg)
for sec in range(0, host_startup_timeout+3):
# hack to ensure that directory is up to date
# ndlist = os.listdir(my_host_invoke.network_dir)
# if verbose: print(str(ndlist))
if os.path.exists(sg):
break
if os.path.exists(prm.abort_path):
log.info('saw abort file %s, aborting test' % prm.abort_path)
break
time.sleep(1)
if not os.path.exists(sg):
abort_test(prm.abort_path, thread_list)
raise Exception('starting signal not seen within %d seconds'
% host_startup_timeout)
if verbose:
print('starting test on host ' + host + ' in 2 seconds')
time.sleep(2 + random.random()) # let other hosts see starting gate file
# FIXME: don't timeout the test,
# instead check thread progress and abort if you see any of them stalled
# but if servers are heavily loaded you can't rely on filesystem
# wait for all threads on this host to finish
for t in thread_list:
my_log.debug('waiting for thread %s' % t.invoke.tid)
t.retrieve()
t.join()
# if not a slave of some other host, print results (for this host)
if not prm_slave:
try:
worker_list = [ t.invoke for t in thread_list ]
output_results.output_results(prm, worker_list)
except FsDriftException as e:
print('ERROR: ' + str(e))
return NOTOK
else:
# if we are participating in a multi-host test
# then write out this host's result in pickle format
# so test driver can pick up result
result_filename = host_result_filename(prm, prm.as_host)
my_log.debug('saving result to filename %s' % result_filename)
worker_list = [ t.invoke for t in thread_list ]
sync_files.write_pickle(result_filename, worker_list)
time.sleep(1.2) # for benefit of NFS with actimeo=1
return OK
def run_multi_host_workload(prm):
prm_host_set = prm.host_set
prm_slave = prm.is_slave
prm_permute_host_dirs = prm.permute_host_dirs
master_invoke = prm.master_invoke
starting_gate = master_invoke.starting_gate
verbose = master_invoke.verbose
host = master_invoke.onhost
# construct list of ssh threads to invoke in parallel
sync_files.create_top_dirs(master_invoke, True)
pickle_fn = os.path.join(prm.master_invoke.network_dir, 'param.pickle')
#if verbose: print('writing ' + pickle_fn)
sync_files.write_pickle(pickle_fn, prm)
if os.getenv('PYPY'):
python_prog = os.getenv('PYPY')
elif sys.version.startswith('2'):
python_prog = 'python'
elif sys.version.startswith('3'):
python_prog = 'python3'
else:
raise Exception('unrecognized python version %s' % sys.version)
#print('python_prog = %s'%python_prog)
remote_thread_list = []
host_ct = len(prm_host_set)
for j in range(0, len(prm_host_set)):
remote_host = prm_host_set[j]
smf_remote_pgm = os.path.join(prm.remote_pgm_dir,
'smallfile_remote.py')
this_remote_cmd = '%s %s --network-sync-dir %s '%\
(python_prog, smf_remote_pgm, prm.master_invoke.network_dir)
#this_remote_cmd = remote_cmd
if prm_permute_host_dirs:
this_remote_cmd += ' --as-host %s' % prm_host_set[(j + 1) %
host_ct]
else:
this_remote_cmd += ' --as-host %s' % remote_host
if verbose: print(this_remote_cmd)
if smallfile.is_windows_os:
remote_thread_list.append(
launcher_thread.launcher_thread(prm, remote_host,
this_remote_cmd))
else:
remote_thread_list.append(
ssh_thread.ssh_thread(remote_host, this_remote_cmd))
# start them, pacing starts so that we don't get ssh errors
for t in remote_thread_list:
t.start()
# wait for hosts to arrive at starting gate
# if only one host, then no wait will occur as starting gate file is already present
# every second we resume scan from last host file not found
# FIXME: for very large host sets, timeout only if no host responds within X seconds
exception_seen = None
hosts_ready = False # set scope outside while loop
abortfn = master_invoke.abort_fn()
last_host_seen = -1
sec = 0
sec_delta = 0.5
host_timeout = prm.host_startup_timeout
if smallfile.is_windows_os: host_timeout += 20
try:
# FIXME: make timeout criteria be that new new hosts seen in X seconds
while sec < host_timeout:
ndirlist = os.listdir(master_invoke.network_dir)
if master_invoke.verbose:
print('shared dir list: ' + str(ndirlist))
hosts_ready = True
if os.path.exists(abortfn):
raise Exception('worker host signaled abort')
for j in range(last_host_seen + 1, len(prm_host_set)):
h = prm_host_set[j]
fn = master_invoke.gen_host_ready_fname(h.strip())
if verbose: print('checking for host filename ' + fn)
if not os.path.exists(fn):
hosts_ready = False
break
last_host_seen = j
if hosts_ready: break
# be patient for large tests
# give user some feedback about how many hosts have arrived at the starting gate
time.sleep(sec_delta)
sec += sec_delta
sec_delta += 1
if verbose:
print('last_host_seen=%d sec=%d' % (last_host_seen, sec))
except KeyboardInterrupt as e:
print('saw SIGINT signal, aborting test')
exception_seen = e
except Exception as e:
exception_seen = e
hosts_ready = False
if not hosts_ready:
smallfile.abort_test(abortfn, [])
if not exception_seen:
raise Exception(
'hosts did not reach starting gate within %d seconds' %
host_timeout)
else:
print('saw exception %s, aborting test' % str(e))
else:
# ask all hosts to start the test
# this is like firing the gun at the track meet
try:
sync_files.write_sync_file(starting_gate, 'hi')
if verbose: print('starting gate file %s created' % starting_gate)
except IOError as e:
print('error writing starting gate: %s' % os.strerror(e.errno))
# wait for them to finish
all_ok = True
for t in remote_thread_list:
t.join()
if t.status != OK:
all_ok = False
print('ERROR: ssh thread for host %s completed with status %d' %
(t.remote_host, t.status))
# attempt to aggregate results by reading pickle files
# containing smf_invocation instances with counters and times that we need
try:
invoke_list = []
for h in prm_host_set: # for each host in test
# read results for each thread run in that host
# from python pickle of the list of smf_invocation objects
pickle_fn = master_invoke.host_result_filename(h)
if verbose: print('reading pickle file: %s' % pickle_fn)
host_invoke_list = []
try:
if not os.path.exists(pickle_fn): time.sleep(1.2)
with open(pickle_fn, 'rb') as pickle_file:
host_invoke_list = pickle.load(pickle_file)
if verbose:
print(' read %d invoke objects' % len(host_invoke_list))
invoke_list.extend(host_invoke_list)
ensure_deleted(pickle_fn)
except IOError as e:
if e.errno != errno.ENOENT: raise e
print(' pickle file %s not found' % pickle_fn)
output_results.output_results(invoke_list, prm_host_set,
prm.thread_count, pct_files_min)
except IOError as e:
print('host %s filename %s: %s' % (h, pickle_fn, str(e)))
all_ok = False
except KeyboardInterrupt as e:
print('control-C signal seen (SIGINT)')
all_ok = False
if not all_ok:
sys.exit(NOTOK)
sys.exit(OK)